
We provide free, open-source software to simulate public policy. Follow our country accounts too: @PolicyEngineUS and @PolicyEngineUK
- Tweets756
- Following162
- Followers1.8K
- Likes876

ALT This speaker card shows moderator Cynthia Osborne of Vanderbilt University and speakers Tara Watson of the Brookings Institution, Max Ghenis of PolicyEngine and Linda Giannarelli of the Urban Institute. They hold the #2026APPAM roundtable "State Policy Impacts on Family Resources: Tools to Improve Policy Analysis." They meet at 10:15 AM ET on Thursday, November 5, in the Wellesley room, 3rd floor, Boston Marriott Copley Place.

ALT PolicyBench weights tax and benefit answers by household impact and ranks models by the share within $1 of PolicyEngine's reference. The chart shows the top 8 of 51 in rank order: GPT-6 Sol 95.2%, Claude Opus 5.5 94.1%, GPT-6.1 Terra 93.1%, GPT-5.6 Sol 93.1%, GPT-6 Astra 91.7%, Claude Sonnet 5.5 91.7%, GPT-6 Luna 91.5%, Claude Fable 5.1 90.8%. A teal bar and a "new" chip mark GPT-6.1 Terra. Models use no tools, and GPT-6.1 Terra exceeds GPT-6.1 Sol's 90.2%.

ALT PolicyBench leaderboard, September 30: share of tax and benefit answers within $1, no tools. 1 GPT-6 Sol 95.0%, 2 Claude Opus 5.5 93.7%, 3 GPT-5.6 Sol 93.6%, 4 GPT-6 Luna 92.1%, 5 Claude Sonnet 5.5 92.1% (new), 6 GPT-6 Astra 91.7%, 7 Claude Fable 5.1 90.8%, 8 GPT-6.1 Sol 90.6% (new), 12 Grok 4.7 88.3% (new), 16 DeepSeek V4.1 Flash 87.1% (new), of 46 models.













